About 20 luminaries of the artificial intelligence (AI) world met in St. Thomas, US Virgin Islands in April 2002 to discuss how best to move the field ahead. This article is a summary of their findings and proceedings, and lays out potential paths for future research and development.

The problem they were working on was that researchers have retreated from the “ambitious aim” of building machines that exhibit “common sense.” They are not now seeking a “unified theory” of AI, but are trying to “develop ways to combine the advantages of multiple methods to represent knowledge, multiple ways to make inferences, and multiple ways to learn” (page 113).

In order to help focus the discussion, Marvin Minsky offered the casual diversity matrix, containing nine cells showing where various AI techniques fit on axes, showing cause and effect. Push Singh, then a graduate student at MIT, suggested a physical model, emphasizing such factors as spatial, physical, and body reasonings, augmented by psychological, social, and reflective characteristics. Ken Forbus noted the role of video game communities in producing virtual worlds, as well as the challenge of robot contests to motivate progress. Aaron Sloman suggested a collection of “mini-scenarios” (page 116) that could be used to test the capabilities of various systems and show progress. After long debates, Minsky and Sloman produced similar architectures for human intelligence systems. “These architectures are distinguished by their emphasis on reflective thinking” (page 118).

The meeting ended with a call for collaboration to continue to build out the architecture. Toward that end, the authors reported that the Defense Advanced Research Projects Agency (DARPA) asked for proposals for building AI systems that contain many of the characteristics discussed at the symposium. It would be interesting to know what progress has been made since.